Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.lang.prolog > #15219

Help! An AI found a Heuristic for me (Was: Good Morning, Vietnam! Sudoku 2026 Challenge)

From Mild Shock <janburse@fastmail.fm>
Newsgroups comp.lang.prolog
Subject Help! An AI found a Heuristic for me (Was: Good Morning, Vietnam! Sudoku 2026 Challenge)
Date 2026-01-10 02:38 +0100
Message-ID <10jsai2$32at2$1@solani.org> (permalink)
References <10jg69u$2nfdv$1@solani.org>

Show all headers | View raw


Hi,

No joke, I have a new CLP heuristic coded,
and made a blind test. Explaind to the AI
what I coded, but not how coded it,

omitted the crucial scoring function, and
the AI was exactly proposing this scoring
funcion out of the blue.

I was quite bamboozled. The AI could not tell
me how it recalled or halucinated the formula,
I spent like 1-2 hours trying to find a

precedent paper. Will blog about it later.
So whats going on . Why can AI suddently
propose Heuristics. Because they are now

differently trained:

Phase 1: Early LLMs (~2018–2020)
Phase 2: LLMs with Prompted Reasoning (~2020–2022)
Phase 3: Scaling & Reasoning Engineering (~2023–2025)

Training an AI to look for heuristics,
can help in self programming. The generative
part of the AI would generate some code,

which then the AI would use for itself. See also:

How it started:

AI for Low-Code for AI
Nikitha Rao, Jason Tsay, Kiran Kate,
Vincent J. Hellendoorn, Martin Hirzel - 2023
https://arxiv.org/abs/2305.20015

How its going:

Successor-Generator Planning with LLM-generated Heuristics
Alexander Tuisov1, Yonatan Vernik2,
Alexander Shleyfman - 2024
https://arxiv.org/abs/2501.18784v4

Bye

Mild Shock schrieb:
> Hi,
> 
> Good Morning Vietnam, the HPC-AI Convergence
> doesn't sleep. Here a friendly reminder of
> the Sudoku leader board (Topn87 Challenge):
> 
> #1: jczsolve / Rust WASM
>      Solving 87 Sudokus in 0.006 seconds (0 sec / Sudoku)
>      https://emerentius.github.io/sudoku_web/
> 
> #2: Kudoku / JavaScript
>      Solving 87 Sudokus in 0.043 seconds (0.0004 sec / Sudoku)
>      https://attractivechaos.github.io/plb/kudoku.html
> 
> #3: Picat / import cp. solve([ff],L)
>      CPU time 0.175 seconds
>      https://picat-lang.org/
> 
> #4: Picat / import sat. solve(L)
>      CPU time 0.373 seconds
>      https://fmv.jku.at/kissat/
> 
> Tested on Windows 11, with a AMD Ryzen AI 350
> 
> Didn't test yet GNU Prolog, ECLiPSe Prolog or
> Ciao Prolog. So whats next? Well beat #1 by
> tapping into an NPU of Copilot+ PC.
> 
> Have Fun!
> 
> Its Winner Winner Chicken Dinner time again...
> 
> Bye
> 

Back to comp.lang.prolog | Previous | NextPrevious in thread | Next in thread | Find similar | Unroll thread


Thread

Good Morning, Vietnam! Sudoku 2026 Challenge Mild Shock <janburse@fastmail.fm> - 2026-01-05 12:11 +0100
  Help! An AI found a Heuristic for me (Was: Good Morning, Vietnam! Sudoku 2026 Challenge) Mild Shock <janburse@fastmail.fm> - 2026-01-10 02:38 +0100
    Failed coping with AI: Moshe Vardi (Was: Help! An AI found a Heuristic for me) Mild Shock <janburse@fastmail.fm> - 2026-01-10 03:02 +0100
      A new proposal: The Gladiator Test (Was: Failed coping with AI: Moshe Vardi) Mild Shock <janburse@fastmail.fm> - 2026-01-10 03:20 +0100
  Chad Brown's Megalodon goes LLM (Was: Good Morning, Vietnam! Sudoku 2026 Challenge) Mild Shock <janburse@fastmail.fm> - 2026-01-10 17:19 +0100
    Magic square of squares [Richard Guy] (Re: Chad Brown's Megalodon goes LLM) Mild Shock <janburse@fastmail.fm> - 2026-01-22 01:40 +0100

csiph-web